We declined to include other imputation methods in this analysis because the Genetic Association Information Network (GAIN) Imputation Working Group is planning to publish a similar comparison using a broad cross-section of methods; our goal here is mainly to benchmark a new method. In order to speed up the analysis via our parallel computing facilities, we split chromosome 10 into 20 non-overlapping analysis chunks. Each imputation run spanned 7 Mb, with an additional 250 kb buffer on either side that was used for inference but omitted from the results – this buffer guards against a deterioration of imputation quality near the chunk edges. We ran every algorithm using the same analysis chunks.
